Acupoint recognition method and acupoint recognition network

Abstract

The application provides an acupoint recognition method and an acupoint recognition network, and relates to the technical field of image recognition, and comprises the following steps: inputting an image to be recognized into a pre-trained feature extraction network to obtain a plurality of feature maps output by the feature extraction network; wherein the feature extraction network comprises a plurality of feature map extraction models connected in series; fusing and adding the feature maps to obtain a fused feature map; inputting the fused feature map into a pre-trained prediction network for acupoint prediction to obtain acupoint coordinates and acupoint categories in the image to be recognized. The application can automatically recognize each acupoint coordinate and the corresponding acupoint type in the image to be recognized, and improves the acupoint recognition speed and the recognition accuracy.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to an acupoint recognition method and acupoint recognition network. Background Technology

[0002] Acupoints hold an important position in Traditional Chinese Medicine (TCM) theory and play a vital role in acupuncture, massage, and acupressure. Acupoints often vary depending on body type, and usually require professionals to accurately locate each acupoint. For those without professional training, it is difficult to find the location and corresponding name of each acupoint. Therefore, how to help non-professionals quickly and accurately identify the location and corresponding acupoint type has become an urgent problem to be solved. Summary of the Invention

[0003] In view of this, the purpose of the present invention is to provide an acupoint recognition method and an acupoint recognition network, which can automatically identify the coordinates of each acupoint and the corresponding acupoint type in the image to be recognized, thereby improving the speed and accuracy of acupoint recognition.

[0004] To achieve the above objectives, the technical solutions adopted in the embodiments of the present invention are as follows:

[0005] In a first aspect, embodiments of the present invention provide an acupoint recognition method, comprising: inputting an image to be recognized into a pre-trained feature extraction network to obtain multiple feature maps output by the feature extraction network; wherein, the feature extraction network includes multiple cascaded feature map extraction models; fusing and adding the feature maps to obtain a fused feature map; and inputting the fused feature map into a pre-trained prediction network for acupoint prediction to obtain the coordinates and acupoint categories of each acupoint in the image to be recognized.

[0006] Furthermore, the present invention provides a first possible implementation of the first aspect, wherein the step of inputting the image to be recognized into a pre-trained feature extraction network to obtain multiple feature maps output by the feature extraction network includes: inputting the image to be recognized into a first feature map extraction model so that the first feature map extraction model outputs a first feature map and a second feature map; inputting the first feature map output by the previous feature map extraction model into a subsequent feature map extraction model so that the subsequent feature map extraction model outputs a first feature map and a second feature map, until all feature map extraction models output second feature maps, and using the second feature maps as multiple feature maps output by the feature extraction network.

[0007] Furthermore, this embodiment of the invention provides a second possible implementation of the first aspect, wherein the feature map extraction model includes multiple extraction modules, a residual module, multiple residual sampling modules, and multiple upsampling modules; the multiple extraction modules are connected in series, the output terminal of the extraction module is connected to the input terminal of the residual module, the output terminal of the residual module is connected to the residual sampling module, the multiple residual sampling modules are connected in series, and the output terminal of the residual sampling module is used to output the first feature map; the input terminal of the upsampling module is connected among the residual sampling modules, and the output terminal of the upsampling module is used to output the second feature map.

[0008] Furthermore, this embodiment of the invention provides a third possible implementation of the first aspect, wherein the plurality of extraction modules include a first extraction module to a fourth extraction module, and the plurality of residual sampling modules include a first residual sampling module to a fourth residual sampling module; the input terminal of the first extraction module is used to input the image to be identified or the first feature map, and the input terminal of the first extraction module is also connected to the output terminal of the fourth residual sampling module; the output terminal of the first extraction module is connected to the input terminal of the second extraction module and the output terminal of the third residual sampling module respectively; the output terminal of the second extraction module is connected to the input terminal of the third extraction module and the output terminal of the second residual sampling module respectively; the output terminal of the third extraction module is connected to the input terminal of the fourth extraction module and the output terminal of the first residual sampling module respectively.

[0009] Furthermore, this embodiment of the invention provides a fourth possible implementation of the first aspect, wherein the plurality of upsampling modules include a first upsampling module to a third upsampling module, the first upsampling module and the second upsampling module are used to perform upsampling twice on the feature map output by the second residual sampling module, and the third upsampling module is used to upsampling the feature map output by the third residual sampling module; the second feature map is obtained by adding the feature map output by the second upsampling module, the feature map output by the third upsampling module and the feature map output by the fourth residual sampling module.

[0010] Furthermore, this embodiment of the invention provides a fifth possible implementation of the first aspect, wherein the extraction module includes a residual module, a pooling layer and a residual module connected in sequence.

[0011] Furthermore, this embodiment of the invention provides a sixth possible implementation of the first aspect, wherein the residual sampling module includes a residual module and an upsampling module.

[0012] Furthermore, this embodiment of the invention provides a seventh possible implementation of the first aspect, wherein the residual module includes a ReLU layer, a batchNormal layer, and a convolutional layer connected in sequence.

[0013] Furthermore, this embodiment of the invention provides an eighth possible implementation of the first aspect, wherein the prediction network includes convolutional layers, batchNormal layers, ReLU layers, and linear layers.

[0014] Secondly, embodiments of the present invention also provide an acupoint recognition network, comprising: a feature extraction network, a feature map fusion network, and a prediction network connected in series; the feature extraction network includes multiple connected feature map extraction models; the feature map extraction model includes multiple extraction modules, residual modules, multiple residual sampling modules, and multiple upsampling modules; the feature extraction network is used to extract multiple feature maps from the image to be recognized and output them to the feature map fusion network; the feature map fusion model is used to fuse and add the feature maps to obtain a fused feature map; the prediction network is used to predict acupoints based on the fused feature map to obtain the coordinates and categories of each acupoint.

[0015] This invention provides an acupoint recognition method and acupoint recognition network, comprising: inputting an image to be recognized into a pre-trained feature extraction network to obtain multiple feature maps output by the feature extraction network; wherein the feature extraction network includes multiple cascaded feature map extraction models; fusing and adding the feature maps to obtain a fused feature map; and inputting the fused feature map into a pre-trained prediction network for acupoint prediction to obtain the coordinates and acupoint categories of each acupoint in the image to be recognized. This invention, by first inputting the image to be recognized into a feature extraction network containing multiple feature map extraction models, can obtain multiple feature maps output by the feature extraction network. By fusing and adding the multiple feature maps to obtain a fused feature map, the fused feature map input into the prediction network can simultaneously retain high-level and low-level features. Through a preset network, the coordinates and corresponding acupoint types of each acupoint in the image to be recognized can be automatically identified, improving the speed and accuracy of acupoint recognition.

[0016] Other features and advantages of the embodiments of the present invention will be set forth in the following description, or some features and advantages may be inferred from the description or determined without doubt, or may be learned by practicing the techniques described above in the embodiments of the present invention.

[0017] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0018] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 A flowchart of an acupoint identification method provided by an embodiment of the present invention is shown;

[0020] Figure 2 A schematic diagram of an acupoint recognition network structure provided by an embodiment of the present invention is shown;

[0021] Figure 3 The diagram shows a schematic of a feature map extraction model structure provided by an embodiment of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be described below in conjunction with the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0023] This embodiment provides an acupoint recognition method applied to electronic devices such as computers. The electronic device is equipped with an acupoint recognition network. (See also...) Figure 1 The flowchart shown illustrates the acupoint identification method, which mainly includes the following steps:

[0024] Step S102: Input the image to be recognized into the pre-trained feature extraction network to obtain multiple feature maps output by the feature extraction network.

[0025] The aforementioned feature extraction network comprises multiple cascaded feature map extraction models. A human image is acquired as the image to be recognized. This image is input into a pre-trained feature extraction network, which then extracts feature maps from the image. Since the feature extraction network includes multiple feature extraction models, feature map extraction is performed based on each model, resulting in multiple feature maps output by the feature extraction network. In one specific implementation, the aforementioned feature extraction network may include eight cascaded feature map extraction models.

[0026] The aforementioned acupoint recognition network includes a feature extraction network, a feature map fusion network, and a prediction network. It acquires various pre-collected human images as acupoint image samples, annotates each acupoint image sample to mark the acupoint coordinates and acupoint category of each acupoint image sample, and inputs the annotated acupoint image samples into the acupoint recognition network for network training to obtain the trained feature extraction network and prediction network.

[0027] Step S104: The feature maps are fused and added together to obtain a fused feature map.

[0028] Multiple feature maps output by the feature extraction network can be fused. For example, the add function can be used to add the feature maps together, and the resulting feature map is called the fused feature map.

[0029] Step S106: Input the fused feature map into the pre-trained prediction network to predict acupoints, and obtain the coordinates and categories of each acupoint in the image to be identified.

[0030] The fused feature map obtained by addition and fusion is input into the trained prediction network for acupoint prediction, so that the prediction network outputs the acupoint coordinates and corresponding acupoint categories of each acupoint in the image to be identified.

[0031] The acupoint recognition method provided in this embodiment first inputs the image to be recognized into a feature extraction network containing multiple feature map extraction models, which can obtain multiple feature maps output by the feature extraction network. By fusing and adding the multiple feature maps, a fused feature map is obtained, which allows the fused feature map input into the prediction network to simultaneously retain high-level features and low-level features. Through the preset network, the coordinates of each acupoint in the image to be recognized and the corresponding acupoint type can be automatically identified, thereby improving the acupoint recognition speed and recognition accuracy.

[0032] See also Figure 2 The diagram shows the structure of the acupoint recognition network. The acupoint recognition network includes a feature extraction network 21, a feature map fusion network 22, and a prediction network 23. The feature extraction network includes multiple feature map extraction modules 210 to 21N connected in series.

[0033] In one embodiment, in order to preserve as many high-level and low-level features of the image as possible, this embodiment provides an implementation method in which the image to be identified is input into a pre-trained feature extraction network to obtain multiple feature maps output by the feature extraction network. Specifically, the following steps (1) to (2) can be performed:

[0034] Step (1): Input the image to be recognized into the first feature map extraction model so that the first feature map extraction model outputs the first feature map and the second feature map.

[0035] The aforementioned feature extraction network includes multiple sequentially connected feature map extraction models. Each feature map extraction module includes two output terminals, referred to as the first output terminal and the second output terminal. The first output terminal of the preceding feature map extraction model is connected to the input terminal of the following feature map extraction model.

[0036] like Figure 2 As shown, the image to be recognized is input into the first feature map extraction model 210. Each feature map extraction model can output two feature maps, which are denoted as the first feature map F1 and the second feature map F2 respectively. The feature map extraction model can output the first feature map F1 through the first output terminal and the second feature map F2 through the second output terminal. The last feature map extraction model only outputs the second feature map F2.

[0037] Step (2): Input the first feature map output by the previous feature map extraction model into the next feature map extraction model so that the next feature map extraction model outputs the first feature map and the second feature map, until the second feature map output by all feature map extraction models is obtained, and the second feature map is used as multiple feature maps output by the feature extraction network.

[0038] like Figure 2 As shown, in the process of extracting feature maps by multiple sequentially connected feature map extraction models, each feature map extraction model outputs two feature maps. The first feature map extraction model inputs the first feature map F1 into the next feature map extraction model through the first output terminal. The second feature map F2 output by each feature map extraction model is input into the feature map fusion network 22 for feature map fusion and addition to obtain a fused feature map. The fused feature map F is then input into the prediction network 23 for acupoint prediction.

[0039] In one implementation, the prediction network can be a deep learning network, such as a convolutional neural network. In a specific implementation, the prediction network may include sequentially connected convolutional layers, batchNormal layers (BN layers, i.e., normalized network layers), ReLU layers (ReLU layers, i.e., activation function layers), and linear layers (fully connected layers). This prediction network may employ a heatmap algorithm to predict acupoint coordinates, and its loss function may be the wingloss algorithm.

[0040] In one embodiment, the feature map extraction model may include an Hourglass network. To further improve the accuracy of acupoint prediction results, the feature map extraction model in this embodiment may be an improved Hourglass network, such as the Hourglass+ network.

[0041] The feature map extraction model provided in this embodiment includes multiple extraction modules (also called RPR modules), residual modules (also called Res modules), multiple residual sampling modules (also called RU modules), and multiple upsampling modules (also called UP modules). The multiple extraction modules are connected in series. The output of the extraction module is connected to the input of the residual module, and the output of the residual module is connected to the residual sampling module. The multiple residual sampling modules are connected in series. The output of the residual sampling module is used to output the first feature map. The input of the upsampling module is connected between the residual sampling modules, and the output of the upsampling module is used to output the second feature map.

[0042] Multiple upsampling modules upsample the feature maps after the second-to-last and third-to-last residual sampling modules, respectively. The feature map obtained by adding the upsampled feature maps is denoted as the second feature map, and the feature map input by the last residual sampling module is denoted as the first feature map.

[0043] See also Figure 3 The schematic diagram of the feature map extraction model structure shown in this embodiment includes a first extraction module RPR1 to a fourth extraction module RPR4, a residual module Res, a first residual sampling module RU1 to a fourth residual sampling module RU4, and a first upsampling module UP1 to a third upsampling module UP3.

[0044] The input terminal of the first extraction module RPR1 is used to input the image to be recognized or the first feature map. The input terminal of the first extraction module RPR1 is also connected to the output terminal of the fourth residual sampling module RU4. In the feature extraction network, the input terminal of the first extraction module in the first feature map extraction model is used to input the image to be recognized, while the input terminal of the first extraction module in other feature map extraction models is used to input the first feature map F1. Figure 3 The diagram shows the structure of the feature map extraction models 211 to 21(N-1) in the feature extraction network.

[0045] like Figure 3 As shown, the output of the first extraction module RPR1 is connected to the input of the second extraction module RPR2 and the output of the third residual sampling module RU3, respectively; the output of the second extraction module RPR2 is connected to the input of the third extraction module RPR3 and the output of the second residual sampling module RU2, respectively; and the output of the third extraction module RPR3 is connected to the input of the fourth extraction module RPR4 and the output of the first residual sampling module RU1, respectively.

[0046] The first upsampling module UP1 and the second upsampling module UP2 are used to upsample the feature map output by the second residual sampling module RU2 twice. The third upsampling module UP3 is used to upsample the feature map output by the third residual sampling module RU3. The second feature map is obtained by adding the feature map output by the second upsampling module UP2, the feature map output by the third upsampling module UP3, and the feature map output by the fourth residual sampling module RU4, which have the same dimension.

[0047] Compared to the existing Hourglass network, the feature extraction network in this embodiment can output one more feature map. The extra second feature map is obtained by adding three feature maps: the feature map obtained by upsampling the feature map after the second residual sampling module RU2 twice, the feature map obtained by upsampling the feature map after the third residual sampling module RU3 once, and the feature map output by the fourth residual sampling module RU4.

[0048] In one implementation, the extraction module is mainly used for feature map extraction, such as using an image feature extraction algorithm. In a specific implementation, such as... Figure 3 As shown, the extraction module RPR provided in this embodiment may include a residual module Res, a pooling layer (Pool) connected in sequence, and the residual module Res may be a ResNet residual network; the residual sampling module RU includes a residual module Res and an upsampling module UP connected in sequence; the residual module Res includes a ReLU layer (activation function layer), a batchNormal layer (BN layer) and a convolutional layer (Conv layer) connected in sequence; the upsampling module UP may use a sampling algorithm such as bilinear interpolation to sample the feature map.

[0049] The feature extraction network provided in this embodiment is a nested residual structure. It retains low-level features to a certain extent during backward propagation, while multiple Hourglasses are concatenated to allow high-level features to be propagated backward. This is a multi-scale fusion approach that can effectively improve the final prediction results.

[0050] When performing keypoint detection, all Hourglass networks are connected in series. The low-level and high-level features retained within an Hourglass are only largely preserved within that single Hourglass. The ability of these features to be preserved across Hourglass networks is very limited. However, the Hourglass+ feature extraction network provided in this embodiment fuses internal features through structural changes and outputs them as residuals. The residual features of all Hourglass+ networks are then added together, forming a residual structure at the Hourglass level. This results in the fusion of Hourglass networks, effectively ensuring that all high-level and low-level features retained by all Hourglass networks are passed to the final prediction network. Furthermore, this feature fusion does not incur additional computational costs and can accelerate the convergence speed of the network during training.

[0051] To verify the aforementioned acupoint recognition network, it was trained on images labeled with acupoints. An acupoint recognition experiment was then conducted on the images to be recognized using the trained network. The results were then evaluated. The results show that the acupoint recognition network achieved 84% recognition of acupoints within an AUC of 0.08, while the Hourglass network achieved 82%. The acupoint prediction time of the acupoint recognition network provided in this embodiment is 5% faster than that of the Hourglass network. These results demonstrate that the acupoint recognition network provided in this embodiment has a faster recognition speed and higher recognition accuracy.

[0052] The acupoint recognition method provided in this embodiment improves the hourglass network by fusing and adding the feature maps obtained by the upsampling of the internal residual sampling module, ensuring that the fused feature map input to the prediction network can simultaneously retain high-level and low-level features, thereby improving the acupoint recognition speed and accuracy.

[0053] Corresponding to the acupoint recognition method provided in the above embodiments, this invention provides an acupoint recognition network, such as... Figure 2 As shown, the acupoint recognition network includes: a feature extraction network 21, a feature map fusion network 22, and a prediction network 23 connected in series; the feature extraction network includes multiple feature map extraction models 210 to 21N connected in series; the feature map extraction model includes multiple extraction modules, residual modules, multiple residual sampling modules, and multiple upsampling modules.

[0054] The aforementioned feature extraction network 21 is used to extract multiple feature maps from the image to be identified and output them to the feature map fusion network.

[0055] Feature map fusion model 22 is used to fuse and add the feature maps to obtain a fused feature map.

[0056] Prediction network 23 is used to predict acupoints based on fused feature maps to obtain the coordinates and categories of each acupoint.

[0057] The acupoint recognition network provided in this embodiment first inputs the image to be recognized into a feature extraction network containing multiple feature map extraction models, which can obtain multiple feature maps output by the feature extraction network. By fusing and adding the multiple feature maps, a fused feature map is obtained, which allows the fused feature map input into the prediction network to simultaneously retain high-level features and low-level features. Through the preset network, the coordinates of each acupoint in the image to be recognized and the corresponding acupoint type can be automatically identified, thereby improving the acupoint recognition speed and recognition accuracy.

[0058] In one embodiment, the feature map extraction model in the above-mentioned feature extraction network can output a first feature map and a second feature map; the first feature map output by the previous feature map extraction model is input into the next feature map extraction model so that the next feature map extraction model outputs a first feature map and a second feature map, until the second feature map output by all feature map extraction models is obtained, and the second feature map is used as multiple feature maps output by the feature extraction network.

[0059] In one embodiment, the feature map extraction model includes multiple extraction modules, a residual module, multiple residual sampling modules, and multiple upsampling modules. The multiple extraction modules are connected in series, with the output of the extraction module connected to the input of the residual module, and the output of the residual module connected to the residual sampling module. The multiple residual sampling modules are connected in series, and the output of the residual sampling module is used to output a first feature map. The input of the upsampling module is connected among the residual sampling modules, and the output of the upsampling module is used to output a second feature map.

[0060] In one embodiment, the plurality of extraction modules include a first extraction module to a fourth extraction module, and the plurality of residual sampling modules include a first residual sampling module to a fourth residual sampling module; the input terminal of the first extraction module is used to input the image to be identified or the first feature map, and the input terminal of the first extraction module is also connected to the output terminal of the fourth residual sampling module; the output terminal of the first extraction module is connected to the input terminal of the second extraction module and the output terminal of the third residual sampling module respectively; the output terminal of the second extraction module is connected to the input terminal of the third extraction module and the output terminal of the second residual sampling module respectively; the output terminal of the third extraction module is connected to the input terminal of the fourth extraction module and the output terminal of the first residual sampling module respectively.

[0061] In one embodiment, the plurality of upsampling modules include a first upsampling module to a third upsampling module. The first upsampling module and the second upsampling module are used to perform upsampling twice on the feature map output by the second residual sampling module. The third upsampling module is used to upsampling the feature map output by the third residual sampling module. The second feature map is obtained by adding the feature map output by the second upsampling module, the feature map output by the third upsampling module, and the feature map output by the fourth residual sampling module.

[0062] In one embodiment, the extraction module includes a residual module, a pooling layer, and a residual module connected in sequence.

[0063] In one embodiment, the residual sampling module includes a residual module and an upsampling module.

[0064] In one embodiment, the aforementioned residual module includes a ReLU layer, a batchNormal layer, and a convolutional layer connected in sequence.

[0065] In one implementation, the prediction network includes convolutional layers, batchNormal layers, ReLU layers, and linear layers.

[0066] The acupoint recognition network provided in this embodiment improves the hourglass network by fusing and adding the feature maps obtained by the upsampling of the internal residual sampling module. This ensures that the fused feature map input to the prediction network can simultaneously retain both high-level and low-level features, thereby improving the acupoint recognition speed and accuracy.

[0067] The acupoint recognition network provided in this embodiment has the same implementation principle and technical effects as the aforementioned embodiments. For the sake of brevity, any parts not mentioned in the network embodiment can be referred to the corresponding content in the aforementioned method embodiment.

[0068] This invention provides an electronic device, which includes a processor and a memory. The memory stores a computer program that can run on the processor. When the processor executes the computer program, it implements the steps of the method provided in the above embodiments.

[0069] This invention provides a computer-readable medium storing computer-executable instructions. When these computer-executable instructions are invoked and executed by a processor, they cause the processor to implement the methods described in the above embodiments.

[0070] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the system described above can be referred to the corresponding process in the foregoing embodiments, and will not be repeated here.

[0071] The computer program product of the acupoint recognition method and acupoint recognition network provided in the embodiments of the present invention includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the methods described in the preceding method embodiments. For specific implementation, please refer to the method embodiments, which will not be repeated here.

[0072] Furthermore, in the description of the embodiments of the present invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in the present invention based on the specific circumstances.

[0073] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0074] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0075] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for identifying acupoints, characterized in that, include: The image to be identified is input into a pre-trained feature extraction network to obtain multiple feature maps output by the feature extraction network; wherein, the feature extraction network includes multiple cascaded feature map extraction models; The feature maps are fused and added together to obtain a fused feature map; The fused feature map is input into a pre-trained prediction network to predict acupoints, thereby obtaining the coordinates and categories of each acupoint in the image to be identified. The step of inputting the image to be recognized into a pre-trained feature extraction network to obtain multiple feature maps output by the feature extraction network includes: inputting the image to be recognized into a first feature map extraction model so that the first feature map extraction model outputs a first feature map and a second feature map; inputting the first feature map output by the previous feature map extraction model into a subsequent feature map extraction model so that the subsequent feature map extraction model outputs a first feature map and a second feature map, until the second feature map output by all feature map extraction models is obtained, and using the second feature map as multiple feature maps output by the feature extraction network; The feature map extraction model includes multiple extraction modules, residual modules, multiple residual sampling modules, and multiple upsampling modules. The extraction modules are connected in series, with the output of each extraction module connected to the input of the residual module, and the output of each residual module connected to the residual sampling module. The residual sampling modules are also connected in series, with the output of each residual sampling module used to output the first feature map. The input of each upsampling module is connected among the residual sampling modules, and the output of each upsampling module is used to output the second feature map.

2. The method according to claim 1, characterized in that, The plurality of extraction modules include a first extraction module to a fourth extraction module, and the plurality of residual sampling modules include a first residual sampling module to a fourth residual sampling module; The input terminal of the first extraction module is used to input the image to be identified or the first feature map, and the input terminal of the first extraction module is also connected to the output terminal of the fourth residual sampling module; The output of the first extraction module is connected to the input of the second extraction module and the output of the third residual sampling module, respectively. The output of the second extraction module is connected to the input of the third extraction module and the output of the second residual sampling module, respectively. The output of the third extraction module is connected to the input of the fourth extraction module and the output of the first residual sampling module, respectively.

3. The method according to claim 2, characterized in that, The plurality of upsampling modules include a first upsampling module, a second upsampling module, and a third upsampling module. The first upsampling module and the second upsampling module are used to upsample the feature map output by the second residual sampling module twice, and the third upsampling module is used to upsample the feature map output by the third residual sampling module. The second feature map is obtained by adding the feature map output by the second upsampling module, the feature map output by the third upsampling module, and the feature map output by the fourth residual sampling module.

4. The method according to claim 1, characterized in that, The extraction module includes a residual module, a pooling layer, and a residual module connected in sequence.

5. The method according to claim 1, characterized in that, The residual sampling module includes a residual module and an upsampling module.

6. The method according to claim 1 or any one of claims 4-5, characterized in that, The residual module includes a ReLU layer, a batchNormal layer, and a convolutional layer connected in sequence.

7. The method according to any one of claims 1-5, characterized in that, The prediction network includes convolutional layers, batchNormal layers, ReLU layers, and linear layers.

8. The method according to claim 6, characterized in that, The prediction network includes convolutional layers, batchNormal layers, ReLU layers, and linear layers.

9. An acupoint recognition device, characterized in that, include: The feature extraction network, feature map fusion network, and prediction network are sequentially connected; the feature extraction network includes multiple sequentially connected feature map extraction models. The feature extraction network is used to extract multiple feature maps from the image to be identified and output them to the feature map fusion network; The feature map fusion model is used to fuse and add the feature maps to obtain a fused feature map; The prediction network is used to predict acupoints based on the fused feature map to obtain the coordinates and categories of each acupoint. The feature extraction network is used to input the image to be identified into the first feature map extraction model, so that the first feature map extraction model outputs a first feature map and a second feature map. The first feature map output by the previous feature map extraction model is input into the next feature map extraction model so that the next feature map extraction model outputs the first feature map and the second feature map, until the second feature map output by all feature map extraction models is obtained. The second feature map is used as multiple feature maps output by the feature extraction network. The feature map extraction model includes multiple extraction modules, residual modules, multiple residual sampling modules, and multiple upsampling modules. The extraction modules are connected in series, with the output of each extraction module connected to the input of the residual module, and the output of each residual module connected to the residual sampling module. The residual sampling modules are also connected in series, with the output of each residual sampling module used to output the first feature map. The input of each upsampling module is connected among the residual sampling modules, and the output of each upsampling module is used to output the second feature map.

Images